Abstract: The power of variable loads in the DC distribution network fluctuates greatly. When the converter power transmission margin (which can be characterized by self-adaption droop coefficient) is small, the converter transmission power is easy to reach the limit, and the ability of responding to the change in power flow will be lost. To cope with above mentioned problem, the converter adaptive droop coefficient, network loss, and nodal voltage offset were taken as the optimizing goals of the DC distribution network structure optimization. The objective function was obtained through the membership function fuzzy processing, and by use of Newton-Raphson method the power flow calculation of the DC distribution network, in which the accurate loss model of the converter was taken account, was carried out. Finally, by means of the iteration of binary particle swarm algorithm the DC distribution network frame structure with synthetical optimum of above mentioned indicators was obtained. The correctness and effectiveness of the proposed DC distribution network frame structure are verified by the modified IEEE33 bus DC distribution network.